import gradio as gr
import tensorflow as tf
from tensorflow import keras
from math import sqrt, ceil
from huggingface_hub import from_pretrained_keras
import numpy as np
model = from_pretrained_keras("keras-io/conditional-gan")
latent_dim = 128
def generate_latent_points(digit, latent_dim, n_samples, n_classes=10):
# generate points in the latent space
random_latent_vectors = tf.random.normal(shape=(n_samples, latent_dim))
labels = tf.keras.utils.to_categorical([digit for _ in range(n_samples)], n_classes)
return tf.concat([random_latent_vectors, labels], 1)
def create_digit_samples(digit, n_samples):
latent_dim = 128
random_vector_labels = generate_latent_points(int(digit), latent_dim, int(n_samples))
examples = model.predict(random_vector_labels)
examples = examples * 255.0
size = ceil(sqrt(n_samples))
digit_images = np.zeros((28*size, 28*size), dtype=float)
n = 0
for i in range(size):
for j in range(size):
if n == n_samples:
break
digit_images[i* 28 : (i+1)*28, j*28 : (j+1)*28] = examples[n, :, :, 0]
n += 1
digit_images = (digit_images/127.5) -1
return digit_images
description = "Keras implementation for Conditional GAN to generate samples for specific digit of MNIST"
article = "Author: Rajeshwar Rathi; Based on the keras example by Sayak Paul"
title = "Conditional GAN for MNIST"
examples = [[1, 10], [3, 5], [5, 15]]
iface = gr.Interface(
fn = create_digit_samples,
inputs = ["number", "number"],
outputs = ["image"],
examples = examples,
description = description,
title = title,
article = article
)
iface.launch()